You are one of the few in the world that would characterize these SOCs as mediocre. Remember, these SOCs are in laptops, and designed to sip power.
So far, the M1 Max is generally quartering the time required for GPU compute tasks in benchmarks for Octane and Redshift (Compared to the M1). This means that they scale well, and puts them in range of a GTX 1080Ti (for GPU rendering). If the rumoured 64- and 128-core variants appear next year (in e.g. the Mac mini, iMac or Mac pro), this family of SOCs will be meeting or exceeding the best coming from individual discrete GPUs.
What's truly interesting are the early results showing their advantages in large scenes, because of the unified memory architecture, bandwidth, and system memory pool. Users have rendered the Disney Moana scene in Redshift in ~26 minutes, versus 46 minutes on a GTX 8000. If these results hold up, it will demonstrate the value of a different path forward for system design...